Breeding park short-term load prediction method, device and equipment and storage medium

Abstract

The application provides a breeding garden short-term load prediction method, device, equipment and storage medium, and relates to the technical field of power load prediction. The method comprises the following steps: obtaining historical monitoring data of a breeding garden to be predicted; based on a grey correlation analysis method, screening features associated with historical power load data from historical environmental data and historical growth characteristic data of breeding animals to obtain a feature reconstruction data set; performing adaptive noise complete set empirical mode decomposition and multivariate variation mode decomposition on the historical power load data to obtain multi-order intrinsic mode load components of the historical power load data; determining effective load components according to the grey correlation degree between the multi-order intrinsic mode load components and the historical power load data; splicing the feature reconstruction data set and the effective load components, and inputting the spliced features into a pre-trained load prediction model to obtain a short-term power load prediction curve of the breeding garden to be predicted. The application can improve the prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of power load forecasting technology, and in particular to a method, apparatus, equipment and storage medium for short-term load forecasting of aquaculture parks. Background Technology

[0002] Large-scale farming is currently developing towards automation and intelligence, with various supporting equipment such as temperature control, ventilation, and feeding systems being put into use on a large scale.

[0003] The electricity load of livestock farms is affected by multiple factors, including seasonal changes, temperature variations, and livestock growth cycles, resulting in significant fluctuations in overall electricity demand, pronounced randomness, and clear load variation patterns. Therefore, achieving high-precision electricity load forecasting is of great value in ensuring the stability of power supply to livestock farms and reducing energy consumption and operating costs in livestock production.

[0004] Short-term load forecasting is a core foundation for power system dispatching and operation, grid economic optimization, and rational allocation of distribution network resources, and it has significant implications for safe power supply and energy-saving dispatching in industrial parks. However, current short-term load forecasting is based on only a single data point, resulting in poor forecast accuracy. Summary of the Invention

[0005] This invention provides a method, apparatus, equipment, and storage medium for short-term load forecasting in aquaculture parks to address the problem of low forecasting accuracy in current methods.

[0006] In a first aspect, embodiments of the present invention provide a method for short-term load forecasting in aquaculture parks, comprising: Obtain historical monitoring data of the breeding park to be predicted; wherein, the historical monitoring data includes at least historical power load data, historical environmental data, and historical growth characteristic data of the farmed animals; Based on grey relational analysis, features associated with the historical power load data are selected from the historical environmental data and the historical growth characteristic data of the farmed animals to obtain a feature reconstruction dataset; Adaptive noise complete ensemble empirical mode decomposition and multivariate variational mode decomposition are performed on the historical power load data to obtain the multi-order intrinsic mode load components of the historical power load data. The effective load component is determined based on the grey correlation between the multi-order intrinsic mode load components and the historical power load data; The reconstructed feature dataset and the effective load component are concatenated, and the concatenated features are input into a pre-trained load prediction model to obtain the short-term power load prediction curve of the aquaculture park to be predicted.

[0007] In one possible implementation, the historical power load data includes a historical active power sequence and a historical reactive power sequence; The process of performing adaptive noisy complete ensemble empirical mode decomposition and multivariate variational mode decomposition on the historical power load data to obtain the intrinsic mode load components of the historical power load data includes: Adaptive noise complete set empirical mode decomposition is performed on the historical active power sequence and the historical reactive power sequence respectively to obtain the eigenmode functions of the high-frequency, mid-frequency and low-frequency components of active power, as well as the eigenmode functions of the high-frequency, mid-frequency and low-frequency components of reactive power. Multivariate variational mode decomposition is performed on the eigenmode functions of the high-frequency components of the active power and the eigenmode functions of the high-frequency components of the reactive power to obtain multiple frequency-stable eigenmode load components of the historical power load data.

[0008] In one possible implementation, before performing adaptive noise-complete set empirical mode decomposition on the historical active power sequence and the historical reactive power sequence, the method further includes: Multiple sets of adaptive white noise are added to the historical active power sequence and the historical reactive power sequence respectively to form multiple noisy historical active power sequences and historical reactive power sequences. The adaptive noise complete set empirical mode decomposition is performed on the historical active power sequence and the historical reactive power sequence, respectively, including... For each noisy historical active power sequence and historical reactive power sequence, perform adaptive noise complete set empirical mode decomposition.

[0009] In one possible implementation, the number of intrinsic mode components and the penalty coefficient in the multivariate variational mode decomposition are determined based on an improved dream optimization algorithm, with the fitness function constructed according to the historical growth characteristics data of the farmed animals and the historical power load data being minimized as the optimization objective.

[0010] In one possible implementation, the historical growth characteristic data of the farmed animals includes the growth cycle and the species of farmed animals; The fitness function is constructed based on the proportions of high-frequency, mid-frequency, and low-frequency components of the historical power load data under each intrinsic modal component, as well as the component complexity, concentration penalty, and load reconfiguration error of the historical power load data under each intrinsic modal component, and their respective weightings; the proportions of high-frequency, mid-frequency, and low-frequency components are determined based on the growth cycle of the aquaculture species. The frequency band range corresponding to the proportions of high-frequency, mid-frequency, and low-frequency components in each intrinsic mode of the historical power load data is determined based on the historical growth characteristics data of the farmed animals and the load fluctuation characteristics of the farm to be predicted.

[0011] In one possible implementation, determining the effective load component based on the grey correlation degree between the multi-order intrinsic mode load components and the historical power load data includes: The grey relational algorithm is used to determine the grey relational degree between the target order intrinsic mode load component and the historical active power sequence and the historical reactive power sequence, respectively; wherein, the historical power load data includes the historical active power sequence and the historical reactive power sequence, and the target order intrinsic mode load component is any one of the multi-order intrinsic mode load components; The average of the two obtained grey correlation degrees is taken as the final correlation degree of the target order intrinsic mode load component; Based on the final correlation degree of all the target order intrinsic mode load components, the intrinsic mode load components that are greater than or equal to the preset correlation degree threshold are superimposed and reconstructed to obtain the effective load components.

[0012] In one possible implementation, the load forecasting model is composed of a Transformer layer and a BiLSTM layer connected in series; The number of attention heads, hidden layer dimensions, and encoder layers in the Transformer layer, as well as the number of neurons, hidden layer dimensions, stacked layers, learning rate, and iteration batch size in the BiLSTM layer, are all determined based on the improved dream optimization algorithm. The method based on grey relational analysis, prior to the historical environmental data and the historical growth characteristic data of the farmed animals, also includes: The historical monitoring data are sequentially processed by filling in missing values, correcting outliers, and normalizing.

[0013] Secondly, embodiments of the present invention provide a short-term load forecasting device for aquaculture parks, comprising: The acquisition module is used to acquire historical monitoring data of the breeding park to be predicted; wherein, the historical monitoring data includes at least historical power load data, historical environmental data, and historical growth characteristic data of the farmed animals; The filtering module is used to filter features associated with the historical power load data from the historical environmental data and the historical growth characteristic data of the farmed animals based on the grey relational analysis method, so as to obtain a feature reconstruction dataset. The decomposition module is used to perform adaptive noise complete set empirical mode decomposition and multivariate variational mode decomposition on the historical power load data to obtain the multi-order intrinsic mode load components of the historical power load data. The determination module is used to determine the effective load component based on the gray correlation degree between the multi-order intrinsic mode load components and the historical power load data; The prediction module is used to concatenate the feature reconstruction dataset and the effective load component, and input the concatenated features into a pre-trained load prediction model to obtain the short-term power load prediction curve of the aquaculture park to be predicted.

[0014] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0015] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0016] Fifthly, embodiments of the present invention provide a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation thereof.

[0017] In this embodiment of the invention, to accurately predict the short-term power load of the aquaculture park, a grey relational analysis method is used to filter out features associated with historical power load data from historical environmental data and historical growth characteristic data of farmed animals, reconstructing a feature reconstruction dataset. To further filter out features related to power load prediction from the historical power load data, adaptive noise complete ensemble empirical mode decomposition and multivariate variational mode decomposition are performed on the historical power load data to obtain multi-order intrinsic mode load components. This allows for the extraction of time-frequency characteristics of the power load, facilitating accurate power load prediction based on these multi-order intrinsic mode load components. Furthermore, by determining the effective load component based on the grey relational degree between the multi-order intrinsic mode load components and the historical power load data, the effective load component can be further determined, accurately distinguishing the real load signal from random noise and eliminating low-correlation noise components. Finally, by concatenating the feature reconstruction dataset and the effective load components, and inputting the concatenated features into a pre-trained load prediction model, the short-term power load prediction curve for the aquaculture park can be obtained. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the implementation of the short-term load forecasting method for aquaculture parks provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the structure of the short-term load forecasting device for aquaculture parks provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0020] As introduced in the background section, short-term load forecasting is the core foundation for power system dispatching and operation, grid economic optimization, and rational allocation of distribution network resources. It has important decision-making significance for safe power supply and energy-saving dispatching in industrial parks.

[0021] The electricity load of livestock farms is affected by multiple factors, including seasonal changes, ambient temperature fluctuations, and the growth cycle of farmed animals, resulting in significant fluctuations, randomness, and volatility in electricity demand. Therefore, accurately predicting the electricity load of livestock farms is the technical problem this invention aims to solve.

[0022] See Figure 1 The document illustrates a flowchart of the implementation of the short-term load forecasting method for aquaculture parks provided in an embodiment of the present invention, which is described in detail below: S110. Obtain historical monitoring data of the aquaculture park to be predicted.

[0023] The historical monitoring data includes at least historical power load data, historical environmental data, and historical growth characteristics data of farmed animals.

[0024] In this embodiment, historical power load data includes historical active power sequences and historical reactive power sequences. Historical environmental data includes temperature, humidity, and air pressure. Historical growth characteristic data of farmed animals includes growth cycles and farmed species, and all historical monitoring data includes time-series characteristics.

[0025] For example, historical monitoring data of the aquaculture park to be predicted can be collected at 15-minute sampling intervals.

[0026] In some embodiments, after obtaining historical monitoring data, in order to facilitate subsequent data processing, the historical monitoring data also needs to be processed sequentially by filling missing values, correcting outliers, and normalizing.

[0027] In this embodiment, for the first Feature sequences If missing points exist Linear interpolation is used to fill in the gaps. ; in: and These represent the nearest non-missing moments before and after the missing point.

[0028] In this embodiment, the following can be used: The principle is to identify outliers and correct them. For any sequence, calculate the mean of the sequence. with standard deviation ,like Then determine If the value is an outlier, replace it with the average of the two normal values ​​before and after that point. ; In this embodiment, to eliminate the influence of different dimensions, min-max normalization is used to map each sequence to... Interval: ; in: The data is after normalization; For the first The original data of the feature sequences; For the first The minimum value of the data in a sequence; For the first The maximum value of data in a sequence.

[0029] In addition, for time attributes such as seasons, one-hot encoding can be used to convert them into numerical vectors, or the seasons can be encoded using a periodic function: ; After completing the above preprocessing, a standardized complete dataset is obtained.

[0030] S120. Based on grey relational analysis, features associated with historical power load data are selected from historical environmental data and historical growth characteristic data of farmed animals to obtain a feature reconstruction dataset.

[0031] The feature reconstruction dataset includes filtered historical environmental data and historical growth characteristic data of farmed animals, and all of these data include time series data.

[0032] In this embodiment, it is assumed that the standardized complete dataset contains There are 10 candidate feature variables, which do not include historical electricity load data. Let the 10th... Feature sequences Historical power load sequence ,in This represents the number of sampling points. It should be noted that historical environmental data, historical growth characteristic data of farmed animals, and historical power load data have all been converted into corresponding sequences.

[0033] For the The characteristic variable at the th ... Grey relational coefficient of each sampling point for: ; ; in: The reference sequence, i.e., the preprocessed original historical power load sequence, is in the [number]th [section]. The normalized value of each sampling point; For comparison sequences, i.e., the first... The candidate characteristic variables (historical environmental data or historical growth characteristic data of farmed animals) in the th... The normalized value of each sampling point; The resolution coefficient is set to 0.5. For the first The correlation between each characteristic variable and the historical power load sequence.

[0034] Set correlation threshold ,Will Feature variables retained, eliminated The feature variables, and all the retained features, constitute the feature reconstruction dataset.

[0035] For example, the correlation threshold The value can range from 0.6 to 0.8.

[0036] S130. Perform adaptive noise complete set empirical mode decomposition and multivariate variational mode decomposition on historical power load data to obtain multi-order intrinsic mode load components of historical power load data.

[0037] In some embodiments, adaptive noise-complete ensemble empirical mode decomposition can be performed first on the historical active power sequence and the historical reactive power sequence to obtain the eigenmode functions (EMFs) of the high-frequency, mid-frequency, and low-frequency components of active power, as well as the EMFs of the high-frequency, mid-frequency, and low-frequency components of reactive power. Then, multivariate variational mode decomposition is performed on the EMFs of the high-frequency components of active power and reactive power to obtain multiple frequency-stable EMFs of the historical power load data.

[0038] In this embodiment, before performing adaptive noise complete set empirical mode decomposition, multiple sets of adaptive white noise can be added to the historical active power sequence and the historical reactive power sequence, respectively, to form multiple noisy historical active power sequences and historical reactive power sequences. Then, adaptive noise complete set empirical mode decomposition is performed on each noisy historical active power sequence and historical reactive power sequence.

[0039] Specifically, multiple sets of Gaussian white noise with different amplitudes are added to the dual-channel sequences of historical active power and reactive power, constructing multiple noisy signals together with the original historical active power and reactive power sequences. Adaptive noise-complete ensemble empirical mode decomposition is then performed on each noisy signal. The components are then input into multivariate variational mode decomposition to further suppress mode aliasing. Iterative steps of residual calculation, noise addition, decomposition, and averaging are repeated until the residuals are monotonic functions, yielding multiple frequency-stable intrinsic mode load components.

[0040] Each intrinsic modal load component characterizes the temporal fluctuation characteristics of the original load in a specific frequency band, with the inherent fluctuation frequency as its core parameter. Different frequency bands correspond to the long-term growth trend and intraday cycle changes of the aquaculture load, respectively.

[0041] In some embodiments, the number of intrinsic mode components and the penalty coefficient in the multivariate variational mode decomposition are determined based on an improved dream optimization algorithm with the goal of minimizing a fitness function. This fitness function is constructed based on historical growth characteristic data of farmed animals and historical electricity load data.

[0042] In this embodiment, the fitness function is constructed based on the proportion of high-frequency components, mid-frequency components, and low-frequency components of historical power load data under each intrinsic modal component, as well as the component complexity, concentration penalty term, and load reconfiguration error of historical power load data under each intrinsic modal component, and their respective weightings; the proportion of high-frequency components, mid-frequency components, and low-frequency components is determined based on the growth cycle of the aquaculture species.

[0043] Among them, the frequency band range of the proportion of high-frequency component, medium-frequency component and low-frequency component in each intrinsic mode component of historical power load data is determined based on the historical growth characteristics data of farmed animals and the load fluctuation characteristics of the farm to be predicted.

[0044] The coefficients corresponding to the component complexity, concentration penalty term, and load reconfiguration error of historical power load data under a set number of modes are the complexity penalty weight coefficient, concentration penalty weight coefficient, and reconfiguration error penalty weight coefficient, respectively, and their values ​​are all determined based on the entropy weight method.

[0045] The weights of the high-frequency, mid-frequency, and low-frequency components in historical power load data under each intrinsic mode component are also determined based on the entropy weight method.

[0046] Specifically, the process can begin by performing adaptive noise-complete ensemble empirical mode decomposition (EMD) on the noisy historical active and reactive power sequences to remove noise. Then, using the number of intrinsic mode components (EMCs) and penalty coefficients from the EMD, the denoised sequences are subjected to EMD to obtain multi-order EMCs. Finally, the decomposition results are substituted into the fitness function to calculate the fitness value corresponding to the current parameters.

[0047] By improving the dream optimization algorithm to generate a population, with a maximum number of iterations set to 100, and minimizing the fitness function value as the optimization objective, the parameters that minimize the fitness function value are iteratively searched in the population to determine the number of intrinsic mode components and penalty coefficients in the multivariate variational mode decomposition.

[0048] The fitness function is: ; in: , , These are the weighting coefficients for the proportions of low-frequency components, mid-frequency components, and high-frequency components, respectively. , , This represents the energy proportion of the load component in each frequency band. The component complexity is represented by the center frequency sample entropy or variance. This is a concentration penalty item; This refers to the load reconfiguration error. , , These are the weighting coefficients for component complexity, concentration penalty term, and load reconstruction error, respectively, used to constrain the quality and reconstruction accuracy of modal components. , , It is determined based on the entropy weight method.

[0049] Among them, the component complexity can be obtained by calculating the sample entropy or variance of the center frequency of the load component, which is used to quantify the randomness of the component interference. The complexity of each intrinsic mode component can be calculated by using sample entropy.

[0050] The concentration penalty term is an evaluation index characterizing the degree of fit between the load component energy and the typical growth cycle of farmed animals. Its value is calculated based on the component center frequency and energy distribution. When the concentration is less than a preset threshold, a larger concentration penalty is applied to strengthen the correlation between the decomposed components and the growth pattern of farmed animals. The calculation formula is as follows: ; ; in: For the first The center frequencies of each intrinsic mode component are calculated from the results of multivariable variational mode decomposition. For the first The energy of each component is obtained by integrating the square of the spectral amplitude. The reference frequency is determined based on the main frequency of load fluctuations corresponding to the typical growth cycle of farmed animals. The concentration threshold is determined based on the statistical distribution of concentrations of known well-decomposed samples in historical data. This is the penalty coefficient.

[0051] Load reconstruction error measures the fidelity of the original signal when all intrinsic mode components are superimposed back into the original signal after decomposition, reflecting whether important information has been lost during the decomposition process.

[0052] ; in: It is the original historical power load sequence, which includes the historical active power sequence and the historical reactive power sequence. The original historical power load sequence The intrinsic mode components obtained from the decomposition. The total number of components; This represents the number of sampling points. The smaller this value, the higher the degree to which the decomposition process retains the original information, and the better the reconstruction fidelity.

[0053] For example, for a 42-day growth cycle of broilers, the low-frequency component bandwidth is set to 30–50 days to characterize the long-term low-frequency load variation trend dominated by the animal's growth process. The medium-frequency component bandwidth is set to 20–28 hours, corresponding to the diurnal rhythm characteristics of power consumption in poultry farming. The high-frequency component bandwidth is set to 15 minutes–2 hours to characterize short-term high-frequency transient load fluctuations caused by the start-up and shutdown of farming equipment and disturbances in the indoor environment. The weighting ratio of each bandwidth is as follows: .

[0054] To accommodate the approximately 500-day growth cycle of laying hens, the low-frequency range was adjusted accordingly. The low-frequency component frequency band was set to a reasonable range around 500 days, adapting to the load variation characteristics of the long-term, stable growth of laying hens. The mid-frequency and high-frequency component ranges remained unchanged, but the weighting of each component was adjusted as follows: This highlights the intraday cyclical dominance of the egg-laying hen farming load.

[0055] S140. Determine the effective load components based on the grey correlation degree between the multi-order intrinsic mode load components and historical power load data.

[0056] In some embodiments, a grey relational algorithm can be used first to determine the grey relational degree between the target-order intrinsic mode load component and the historical active power sequence and the historical reactive power sequence, respectively. Then, the average of the two grey relational degrees is taken as the final relational degree of the target-order intrinsic mode load component. Finally, based on the final relational degrees of all target-order intrinsic mode load components, intrinsic mode load components greater than or equal to a preset relational degree threshold are superimposed and reconstructed to obtain the effective load component.

[0057] The historical power load data includes historical active power sequences and historical reactive power sequences, and the target order intrinsic mode load component is any one of the multi-order intrinsic mode load components.

[0058] In this embodiment, the original historical active power sequence and historical reactive power sequence are used as reference sequences, and the intrinsic mode load components obtained from the decomposition are used as comparison sequences. Grey relational analysis is used to calculate the grey relational degree between each intrinsic mode load component and the original historical active power sequence and historical reactive power sequence. For each intrinsic mode load component, its relational degree with the active power sequence is calculated separately. and its correlation with reactive power series The average of the two values ​​is taken as the final correlation degree of the component. The calculation formula is: ; ; In the formula: For the first Grey relational coefficient of each intrinsic mode load component relative to the original historical active power sequence or historical reactive power sequence; The first of the original historical active power sequence or historical reactive power sequence Values ​​of each feature point; For the first The first intrinsic modal load component Values ​​of each feature point; The resolution coefficient is set to 0.5. For the first Grey relational degree between load component and active power P and reactive power Q.

[0059] The intrinsic modal load components are sorted according to the calculated grey relational degree. Components with grey relational degree below the preset threshold are removed, and effective components with high grey relational degree are retained. The effective components are then superimposed and reconstructed to obtain the reconstructed effective load sequence.

[0060] S150. The feature reconstruction dataset and the effective load component are concatenated, and the concatenated features are input into the pre-trained load prediction model to obtain the short-term power load prediction curve of the aquaculture park to be predicted.

[0061] In some embodiments, the load forecasting model is composed of a Transformer layer and a BiLSTM layer connected in series.

[0062] In this embodiment, the number of attention heads, hidden layer dimensions, and encoder layers in the Transformer layer, as well as the number of neurons, hidden layer dimensions, stacked layers, learning rate, and iteration batch size in the BiLSTM layer, are all determined based on the improved dream optimization algorithm.

[0063] Specifically, define the set of hyperparameters to be optimized: ; in, For hyperparameter set; This represents the number of attention heads in the Transformer layer. The hidden layer dimension in the Transformer layer; This represents the number of encoder layers in the Transformer layer. This represents the number of BiLSTM neurons; Let the hidden layer dimension of BiLSTM be denoted as . The number of stacked layers in the BiLSTM; The learning rate; This is the batch size for iteration.

[0064] By improving the dream optimization algorithm to generate a hyperparameter population, and simulating the dream generation, dream mutation, and awakening update mechanism to iteratively optimize, the loss function is obtained. Minimum optimal hyperparameter The optimal hyperparameters are then configured into the hybrid model to obtain the final load prediction model.

[0065] The loss function uses the root mean square error: ; in, This represents the actual load component. To predict load components, This represents the number of samples.

[0066] In some embodiments, the feature reconstruction dataset and the payload components are concatenated to construct a feature matrix: ; in: The effective load component matrix; A matrix of historical growth characteristics data for farmed animals. For environmental data matrix, For time step, This represents the total feature dimension.

[0067] The spliced ​​features The data is input into a pre-trained load prediction model to obtain the predicted load component sequence. : ; Then, for the predicted load component sequence Perform inverse normalization and component superposition. ; in: The number of valid components; For the first The predicted values ​​after inverse normalization of each component are used to obtain the short-term power load forecast curve for the aquaculture park for the next day.

[0068] To accurately predict the short-term power load of a livestock farm, this invention employs grey relational analysis to filter out features associated with historical environmental data, historical growth characteristics of farmed animals, and historical power load data, reconstructing a feature reconstruction dataset. To further filter features relevant to power load prediction from the historical power load data, adaptive noisy complete ensemble empirical mode decomposition and multivariate variational mode decomposition are performed on the historical power load data to obtain multi-order intrinsic mode load components. This allows for the extraction of time-frequency characteristics of the power load, facilitating accurate power load prediction based on these multi-order intrinsic mode load components. Furthermore, by determining the effective load component based on the grey relational degree between the multi-order intrinsic mode load components and the historical power load data, the effective load component can be accurately distinguished from the actual load signal and random noise, eliminating low-correlation noise components. Finally, by concatenating the feature reconstruction dataset and the effective load components, and inputting the concatenated features into a pre-trained load prediction model, the short-term power load prediction curve for the livestock farm can be obtained.

[0069] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0070] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0071] Figure 2A schematic diagram of the short-term load forecasting device for aquaculture parks provided in an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below: like Figure 2 As shown, the short-term load forecasting device for aquaculture parks includes: The acquisition module 210 is used to acquire historical monitoring data of the breeding park to be predicted; wherein, the historical monitoring data includes at least historical power load data, historical environmental data and historical growth characteristic data of the farmed animals; The filtering module 220 is used to filter features associated with historical power load data from historical environmental data and historical growth characteristic data of farmed animals based on grey relational analysis, and obtain a feature reconstruction dataset. The decomposition module 230 is used to perform adaptive noise complete set empirical mode decomposition and multivariate variational mode decomposition on historical power load data to obtain multi-order intrinsic mode load components of historical power load data. The determination module 240 is used to determine the effective load component based on the grey correlation degree between the multi-order intrinsic mode load components and historical power load data; The prediction module 250 is used to concatenate the feature reconstruction dataset and the effective load component, and input the concatenated features into the pre-trained load prediction model to obtain the short-term power load prediction curve of the aquaculture park to be predicted.

[0072] In one possible implementation, historical power load data includes historical active power sequences and historical reactive power sequences. The decomposition module 230 is used to perform adaptive noise complete set empirical mode decomposition on the historical active power sequence and the historical reactive power sequence respectively, to obtain the eigenmode functions of the high-frequency, mid-frequency and low-frequency components of active power, as well as the eigenmode functions of the high-frequency, mid-frequency and low-frequency components of reactive power. Multivariate variational mode decomposition is performed on the eigenmode functions of the high-frequency components of active power and reactive power to obtain multiple frequency-stable eigenmode load components from historical power load data.

[0073] In one possible implementation, the decomposition module 230 is used to add multiple sets of adaptive white noise to the historical active power sequence and the historical reactive power sequence respectively, thereby forming multiple noisy historical active power sequences and historical reactive power sequences. For each noisy historical active power sequence and historical reactive power sequence, perform adaptive noise complete set empirical mode decomposition.

[0074] In one possible implementation, the number of intrinsic mode components and the penalty coefficient in the multivariate variational mode decomposition are determined by minimizing the fitness function constructed based on the historical growth characteristics data and historical power load data of farmed animals, using an improved dream optimization algorithm as the optimization objective.

[0075] In one possible implementation, the historical growth characteristics data of farmed animals include growth cycle and farmed species; The fitness function is constructed based on the proportions of high-frequency, mid-frequency, and low-frequency components in historical power load data under each intrinsic modal component, as well as the component complexity, concentration penalty term, and load reconfiguration error of historical power load data under each intrinsic modal component, and their respective weightings; the proportions of high-frequency, mid-frequency, and low-frequency components are determined based on the growth cycle of the aquaculture species. The frequency band ranges of the proportions of high-frequency, mid-frequency, and low-frequency components in historical power load data under each intrinsic mode component are determined based on historical growth characteristic data of farmed animals and the load fluctuation characteristics of the farm to be predicted.

[0076] In one possible implementation, the determining module 240 is used to determine the grey relational degree between the target order intrinsic mode load component and the historical active power sequence and the historical reactive power sequence, respectively, using a grey relational algorithm; wherein, the historical power load data includes the historical active power sequence and the historical reactive power sequence, and the target order intrinsic mode load component is any one of the multi-order intrinsic mode load components; The average of the two obtained grey correlation degrees is taken as the final correlation degree of the target order intrinsic mode load component; Based on the final correlation degree of all target order intrinsic mode load components, the intrinsic mode load components that are greater than or equal to the preset correlation degree threshold are superimposed and reconstructed to obtain the effective load components.

[0077] In one possible implementation, the load forecasting model is composed of a Transformer layer and a BiLSTM layer connected in series; The number of attention heads, hidden layer dimensions, and encoder layers in the Transformer layer, as well as the number of neurons, hidden layer dimensions, stacked layers, learning rate, and iteration batch size in the BiLSTM layer, are all determined based on the improved dream optimization algorithm.

[0078] Figure 3 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 3As shown, the electronic device 3 in this embodiment includes a processor 30 and a memory 31. The memory 31 stores a computer program 32. When the processor 30 executes the computer program 32, it implements the steps in the various method embodiments described above. Alternatively, when the processor 30 executes the computer program 32, it implements the functions of each module / unit in the various device embodiments described above.

[0079] For example, computer program 32 may be divided into one or more modules / units, which are stored in memory 31 and executed by processor 30 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 32 in electronic device 3.

[0080] Electronic device 3 may include, but is not limited to, processor 30 and memory 31. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 3 may also include input / output devices, network access devices, buses, etc.

[0081] The processor 30 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0082] The memory 31 can be an internal storage unit of the electronic device 3, such as a hard disk or RAM. The memory 31 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card (FC) equipped on the electronic device 3. Furthermore, the memory 31 can include both internal and external storage units of the electronic device 3. The memory 31 is used to store the computer program 32 and other programs and data required by the electronic device 3. The memory 31 can also be used to temporarily store data that has been output or will be output.

[0083] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0084] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0085] This invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0086] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0087] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0088] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for short-term load forecasting in aquaculture parks, characterized in that, include: Obtain historical monitoring data of the breeding park to be predicted; wherein, the historical monitoring data includes at least historical power load data, historical environmental data, and historical growth characteristic data of the farmed animals; Based on grey relational analysis, features associated with the historical power load data are selected from the historical environmental data and the historical growth characteristic data of the farmed animals to obtain a feature reconstruction dataset; Adaptive noise complete ensemble empirical mode decomposition and multivariate variational mode decomposition are performed on the historical power load data to obtain the multi-order intrinsic mode load components of the historical power load data. The effective load component is determined based on the grey correlation between the multi-order intrinsic mode load components and the historical power load data; The reconstructed feature dataset and the effective load component are concatenated, and the concatenated features are input into a pre-trained load prediction model to obtain the short-term power load prediction curve of the aquaculture park to be predicted.

2. The short-term load forecasting method for aquaculture parks according to claim 1, characterized in that, The historical power load data includes historical active power sequences and historical reactive power sequences. The process of performing adaptive noise-complete ensemble empirical mode decomposition and multivariate variational mode decomposition on the historical power load data to obtain the intrinsic mode load components of the historical power load data includes: Adaptive noise complete set empirical mode decomposition is performed on the historical active power sequence and the historical reactive power sequence respectively to obtain the eigenmode functions of the high-frequency, mid-frequency and low-frequency components of active power, as well as the eigenmode functions of the high-frequency, mid-frequency and low-frequency components of reactive power. Multivariate variational mode decomposition is performed on the eigenmode functions of the high-frequency components of the active power and the eigenmode functions of the high-frequency components of the reactive power to obtain multiple frequency-stable eigenmode load components of the historical power load data.

3. The short-term load forecasting method for aquaculture parks according to claim 2, characterized in that, Before performing adaptive noise complete set empirical mode decomposition on the historical active power sequence and the historical reactive power sequence respectively, the method further includes: Multiple sets of adaptive white noise are added to the historical active power sequence and the historical reactive power sequence respectively to form multiple noisy historical active power sequences and historical reactive power sequences. The adaptive noise complete set empirical mode decomposition is performed on the historical active power sequence and the historical reactive power sequence, respectively, including... For each noisy historical active power sequence and historical reactive power sequence, perform adaptive noise complete set empirical mode decomposition.

4. The short-term load forecasting method for aquaculture parks according to claim 1, characterized in that, The number of intrinsic mode components and the penalty coefficient in the multivariate variational mode decomposition are determined based on the improved dream optimization algorithm, which uses the fitness function constructed from the historical growth characteristics data of the farmed animals and the historical power load data as the optimization objective.

5. The short-term load forecasting method for aquaculture parks according to claim 4, characterized in that, The historical growth characteristics data of the farmed animals include growth cycle and farmed species; The fitness function is constructed based on the proportions of high-frequency, mid-frequency, and low-frequency components of the historical power load data under each intrinsic modal component, as well as the component complexity, concentration penalty, and load reconfiguration error of the historical power load data under each intrinsic modal component, and their respective weightings; the proportions of high-frequency, mid-frequency, and low-frequency components are determined based on the growth cycle of the aquaculture species. The frequency band range of the proportion of high-frequency components, mid-frequency components, and low-frequency components in each intrinsic mode component of the historical power load data is determined based on the historical growth characteristic data of the farmed animals and the load fluctuation characteristics of the farm to be predicted.

6. The short-term load forecasting method for aquaculture parks according to any one of claims 1-5, characterized in that, The step of determining the effective load component based on the grey correlation between the multi-order intrinsic mode load components and the historical power load data includes: The grey relational algorithm is used to determine the grey relational degree between the target order intrinsic mode load component and the historical active power sequence and the historical reactive power sequence, respectively; wherein, the historical power load data includes the historical active power sequence and the historical reactive power sequence, and the target order intrinsic mode load component is any one of the multi-order intrinsic mode load components; The average of the two obtained grey correlation degrees is taken as the final correlation degree of the target order intrinsic mode load component; Based on the final correlation degree of all the target order intrinsic mode load components, the intrinsic mode load components that are greater than or equal to the preset correlation degree threshold are superimposed and reconstructed to obtain the effective load components.

7. The short-term load forecasting method for aquaculture parks according to any one of claims 1-5, characterized in that, The load prediction model is composed of a Transformer layer and a BiLSTM layer connected in series; The number of attention heads, hidden layer dimensions, and encoder layers in the Transformer layer, as well as the number of neurons, hidden layer dimensions, stacked layers, learning rate, and iteration batch size in the BiLSTM layer, are all determined based on the improved dream optimization algorithm. The method based on grey relational analysis, prior to the historical environmental data and the historical growth characteristic data of the farmed animals, also includes: The historical monitoring data are sequentially processed by filling in missing values, correcting outliers, and normalizing.

8. A short-term load forecasting device for aquaculture parks, characterized in that, include: The acquisition module is used to acquire historical monitoring data of the breeding park to be predicted; wherein, the historical monitoring data includes at least historical power load data, historical environmental data, and historical growth characteristic data of the farmed animals; The filtering module is used to filter features associated with the historical power load data from the historical environmental data and the historical growth characteristic data of the farmed animals based on the grey relational analysis method, so as to obtain a feature reconstruction dataset. The decomposition module is used to perform adaptive noise complete set empirical mode decomposition and multivariate variational mode decomposition on the historical power load data to obtain the multi-order intrinsic mode load components of the historical power load data. The determination module is used to determine the effective load component based on the gray correlation degree between the multi-order intrinsic mode load components and the historical power load data; The prediction module is used to concatenate the feature reconstruction dataset and the effective load component, and input the concatenated features into a pre-trained load prediction model to obtain the short-term power load prediction curve of the aquaculture park to be predicted.

9. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images